shenxingy/ads-server-side-tracking

Server-Side Tracking Pipeline Audit

Audits the entire server-side measurement pipeline that backs every paid channel's modeled conversion data. Without server-side tracking in 2026, expect 30-40% conversion data loss from iOS ATT, ITP, and aggressive ad blockers — that's the gap between what's actually happening and what your bid algorithms can see.

This sub-skill is technical and deep. It's NOT the same as ads-attribution, which audits the attribution model sitting on top of these events.

Process

1. Collect server-side stack inventory: sGTM container info, Meta CAPI integration method (Gateway / direct / partner integration), event schema documentation, hosting infrastructure (Cloud Run / GCS / AWS)
2. Read ads/references/conversion-tracking.md for cross-platform baseline
3. Test event flow: trigger known events → verify they appear in BOTH client-side (Pixel Helper / Tag Assistant) AND server-side (Events Manager test events / GA4 DebugView)
4. Audit deduplication, hashing, and parameter completeness
5. Score health PASS / WARNING / FAIL per surface
6. Generate findings report

What to Analyze

Server-Side Google Tag Manager (sGTM)

• sGTM container deployed — hosted on Cloud Run, GCS, App Engine, or custom infrastructure. Self-hosted preferred over Google-managed for cost and data residency
• Custom domain configured (tags.example.com) — first-party domain avoids ITP / ad-blocker blocking that hits googletagmanager.com
• Client-side GTM forwards to sGTM correctly; cookies, IP, user-agent preserved
• GA4 events flow via sGTM (no direct client → GA4 fallback)
• Conversion Linker tag enabled — preserves Google click IDs (gclid, gbraid, wbraid) across cross-domain navigation
• Server-side privacy filters — strip non-essential PII before forwarding to analytics; only hash + forward what's needed for matching

Meta CAPI / CAPI Gateway

• CAPI active — Conversions API server-to-server alongside the Pixel
• CAPI Gateway preferred over manual server implementation (auto- hashing, parameter coverage, lower maintenance)
• All major events server-side: PageView, ViewContent, AddToCart, InitiateCheckout, Purchase, Lead, CompleteRegistration
• Event Match Quality (EMQ) ≥8.0 for Purchase — confirm via Events Manager → Overview → Data sources
• customer_information parameters sent server-side: em (email), ph (phone), fn/ln (name), ct/st/zp (geo), external_id, client_ip_address, client_user_agent, fbc, fbp
• Hashing: lowercased + trimmed SHA-256 for PII fields BEFORE send
• action_source field set per event (website, app, physical_store, email, system_generated)

Event Deduplication

• event_id generated client-side, included in BOTH the Pixel event AND the CAPI / sGTM payload — Meta + Google both dedupe on this
• Dedup rate ≥90% measured in Events Manager → Diagnostics
• Timestamp alignment — server-side event timestamp within 5 minutes of client-side counterpart
• event_name consistency — server-side uses the same canonical event names as client-side (don't rename in transit)

Server-Side Hit Ratio

• Server-side ≥80% of client-side hits for Purchase / Lead — anything lower means iOS / ITP / ad-blocker data loss isn't being recovered
• Server-side >100% acceptable — means server-side captures conversions the client-side missed (good — that's what server-side is for)
• Hit ratio monitored over time — drops below 60% indicate broken server-side firing or missing event_id

Pixel / Tag Debug Walkthrough

When deployed, validate every event end-to-end:

• Facebook Pixel Helper (Chrome extension) shows the Pixel firing client-side with correct event_name + event_id + value + currency
• Meta Events Manager → Test Events shows the CAPI event arriving server- side with matching event_id and customer_information parameters populated
• Google Tag Assistant confirms client-side gtag firing
• GA4 DebugView confirms server-side event arriving with event params
• Network tab shows client → sGTM forwarding (not client → Google direct)
• window.dataLayer populates expected variables before any tag fires

Custom Event Taxonomy

• Canonical event names documented (e.g., purchase not Purchase or PURCHASE or order_complete)
• Standard params per event: value, currency, content_ids, content_type, num_items
• Custom params namespaced (cx_segment, cx_funnel_step) to avoid collision with platform-standard params
• Schema versioned — when the taxonomy changes, bump a version param so downstream platforms can handle the cutover

Hash Quality & PII Handling

• Email: lowercased + trimmed + SHA-256 (no other normalization)
• Phone: E.164 format + SHA-256 (e.g., +15551234567)
• Name: lowercased + trimmed + SHA-256 per first / last separately
• City / state / zip: lowercased + SHA-256
• NEVER hash already-hashed values — double-hashing breaks matching
• NEVER send plain PII server-side — only hashed
• GDPR / CPRA / CDPA compliance: confirm consent state is read before sending PII server-side, even hashed

Key Thresholds

| Metric | Pass | Warning | Fail | |--------|------|---------|------| | sGTM custom domain | Active | Configured, not active | Not configured | | CAPI Gateway | Active | Manual CAPI | Pixel-only | | EMQ (Purchase) | ≥8.0 | 6.0-7.9 | <6.0 | | Dedup rate | ≥90% | 70-89% | <70% | | Server / client hit ratio | 80-120% | 50-79% | <50% | | customer_information completeness | 6+ params | 4-5 params | <4 params | | Hash convention | Documented + verified | Implicit | Inconsistent | | Test events validation | All 6 events pass | 3-5 events pass | <3 events pass |

Output

Server-Side Tracking Health Score

Server-Side Tracking Health Score: XX/100 (Grade: X)

sGTM Pipeline:               XX/100  ████████░░  (20%)
CAPI / CAPI Gateway:         XX/100  ██████████  (25%)
Deduplication:               XX/100  █████████░  (15%)
Server-Side Hit Ratio:       XX/100  ████████░░  (15%)
Pixel Debug (6 events):      XX/100  ███████░░░  (10%)
Hash Quality / PII Handling: XX/100  ██████░░░░  (15%)

Deliverables

• SERVER-SIDE-TRACKING-AUDIT.md: Full pipeline findings
• Test-event reproduction log (which events validated end-to-end on which date, with screenshots from Events Manager / DebugView)
• EMQ improvement roadmap (parameter-by-parameter)
• Hit-ratio dashboard recommendation
• Pre-launch checklist for any new platform integration (Amazon Marketing Cloud, Apple Ads, TikTok Events API)